CITADEL
StandbyNothing is live right now.
Eval accuracylast measured Tue 10 Nov, 15:27 UTC
%
0.9127
one eval
0%25%50%75%100%0%25%50%75%100%0%25%50%75%100%

Claviger router: OneJev 0.8B (backfilled)

finishedRadeon 8060S

A model that routes a request to the one destination, of many, that should take it. Backfilled: recorded after the fact from the run's own logs and records. Times inside a run are estimated, and there are no GPU sensor readings.

started
Tue 10 Nov, 14:30 UTC
ran for
45:50
config
learning rate 0.00002, batch 1, seed 0

run2/5step42/42epoch1/1config2/3seed0lr0.00002batch1modelOneJev1.6-0.8B 0.8Brevc89933c

Loss train 0.6162 ▲ +0.5287 eval 0.5240 lowest 0.0001 at step 3 run before, run 1 ▼ −0.4414 step 42
Timebase
Range
Smoothing
Reference
42 readings, 30 min/div10 min/div5 min/div
0123400:0030:00phase starting, 0.0 s to 65.5 sphase training, 65.5 s to 45:50epoch 1, 65.5 straining+0.5287since first0.0876Lowest last train loss across this project's 5 runs, run 4lowest last 0.0014, run 40.61620123400:0010:0020:0030:0040:00phase starting, 0.0 s to 65.5 sphase training, 65.5 s to 45:50epoch 1, 65.5 straining0.0876Lowest last train loss across this project's 5 runs, run 4lowest last 0.0014, run 4eval 0.52400.6162▲ +0.52870123400:0005:0010:0015:0020:0025:0030:0035:0040:0045:00phase starting, 0.0 s to 65.5 sphase training, 65.5 s to 45:50epoch 1, 65.5 straining0.0876Lowest last train loss across this project's 5 runs, run 4lowest last 0.0014, run 4eval 0.52400.6162▲ +0.5287

Eval measures

Eval losslower is better
0.5240
02 →
ECEcalibration error, lower is better
0.0513
01

Elapsed

65.5 s to epoch 1, 2684.9 s to eval 1

42 of 42 steps, 1 of 1 epochs, 100%

65.5 s per step, 0.9 steps a minute

Against the other runs

2 of 3

by last eval accuracy

best so far 0.9233, run 1

▼ −0.0106 against run 1

Train minus eval

+0.0922

step 42

train 0.6162, eval 0.5240

Set-up

666 train

synthetic, learning rate 0.00002, batch size 1, grad accum 16, epochs 1, optimizer adamw, seed 0

Learning rate

0.00e+0

▼ −2.00e-5 since first

Input

  • Runs128 runs
  • Agentsnone showing
  • Treesnone showing
  • Projectsnone showing

Check

Every entry is checked on its way out; some are held for review.

Not shown on any run: data samples, predictions, raw logs, file paths, machine names

Delay

2 min, set

Watching the agents work, about 2 min behind.

One way

Nothing here reaches the agents.

This page

This page updates by itself.

visits counted with Google Analytics · choose

Runsheight is eval accuracyall recordings ▸ | finished× failed■ stopped○ lost● live

2 of 5 finished, 0.9127, config 2 of 3 (learning rate 0.00002, batch 1), seed 0, Tue 10 Nov, 14:30 UTC

Runs as a table
StartedStatusEval accuracyStepsConfig
Tue 10 Nov, 17:10 UTC| finished-42 of 42learning rate 0.00001, batch 1, seed 2 of 3
Tue 10 Nov, 16:24 UTC| finished-42 of 42learning rate 0.00001, batch 1, seed 1 of 3
Tue 10 Nov, 15:27 UTC| finished0.084742 of 42learning rate 0.00005, batch 1, seed 0
Tue 10 Nov, 14:30 UTC| finished0.912742 of 42learning rate 0.00002, batch 1, seed 0
Tue 10 Nov, 13:33 UTC| finished0.923342 of 42learning rate 0.00001, batch 1, seed 0 of 3

Plate

Model
OmniJev/OneJev1.6-0.8B, 0.8B parameters revision c89933c
Data
Synthetic routing worlds, synthetic
Config
learning rate 0.00002, batch size 1, grad accum 16, epochs 1, optimizer adamw, seed 0
Hardware
Radeon 8060S, ROCm

Log

Since startMark
0.0 sphase starting
65.5 sphase training
65.5 sepoch 1
2750.4 seval 1
190 h 22 min laterwent public

Readings

42 rows as a table
StepSecondsTrain lossEval lossEval accuracyECELearning rate
165.50.0876---2.00e-5
2131.00.0181---2.00e-5
3196.50.0001---1.99e-5
4261.91.0324---1.97e-5
5327.40.4853---1.95e-5
6392.90.6367---1.93e-5
7458.42.2979---1.90e-5
8523.92.4386---1.86e-5
9589.40.8418---1.82e-5
10654.90.0428---1.77e-5
11720.30.1511---1.72e-5
12785.82.9916---1.67e-5
13851.30.7542---1.61e-5
14916.81.1569---1.54e-5
15982.31.3440---1.48e-5
161047.81.4812---1.41e-5
171113.30.4683---1.34e-5
181178.70.9359---1.26e-5
191244.21.3779---1.19e-5
201309.70.1052---1.11e-5
211375.20.5356---1.04e-5
221440.71.2837---9.62e-6
231506.20.0911---8.85e-6
241571.70.4536---8.10e-6
251637.10.8095---7.35e-6
261702.60.7306---6.62e-6
271768.10.4166---5.91e-6
281833.60.8921---5.22e-6
291899.10.5972---4.56e-6
301964.60.6265---3.94e-6
312030.10.2932---3.35e-6
322095.50.0231---2.79e-6
332161.00.7083---2.29e-6
342226.50.6833---1.82e-6
352292.00.9929---1.40e-6
362357.50.2946---1.04e-6
372423.00.8930---7.25e-7
382488.51.1404---4.66e-7
392553.90.1858---2.63e-7
402619.40.2482---1.17e-7
412684.90.4858---2.93e-8
422750.40.61620.52400.91270.05130.00e+0

Worked out

ReadoutValueWorked out from
Run complete100%42 steps over 42
Step time65.5 s2619.4 s over 40 steps, between readings. Startup and intervals that hold an eval are left out
Steps per minute0.960 s over the step time of 65.48 s
Train loss since first reading+0.52870.6162 at step 42 minus 0.0876 at step 1
Lowest train loss0.0001Lowest of 42 readings, at step 3
Eval accuracy, as shown91.3%0.9127 times 100, to one decimal. Changes on that plate are percentage points
Train minus eval loss+0.09220.6162 minus 0.5240, both at step 42
Rank2 of 3By last eval accuracy, across this project's 3 runs with an eval
Best so far0.9233Highest last eval accuracy, run 1
Against the run before−0.01060.9127 minus 0.9233 on run 1
Lowest last train loss0.0014Lowest last train loss across this project's 5 runs, run 4
Items per second0.01 step over 65.48 s, times batch size 1
Train loss against run 1−0.44140.6162 minus 1.0576 on run 1, both at step 42
Learning rate since first reading−2.00e-50.00e+0 minus 2.00e-5